Statistical model development

RandomForest_model <- function(input, formula, write=F){
  
  OOB_error_rf <- vector()
  Accuracy_rf  <- vector()
  #create empty table for species
  confusion_mat_spp_sum <- as.data.frame(matrix(ncol = length(unique(input$SPECIES)), nrow = length(unique(input$SPECIES)) ));
  colnames(confusion_mat_spp_sum) <- levels(unlist(input$SPECIES))
  rownames(confusion_mat_spp_sum) <- levels(unlist(input$SPECIES))
  confusion_mat_spp_sum[,] <- 0
  
  #create empty table for softwood hardwood
  # confusion_mat_spp_sum <- as.data.frame(matrix(ncol = 2, nrow = 2));
  # colnames(confusion_mat_spp_sum) <- c("Broadleaf", "Needleleaf")
  # rownames(confusion_mat_spp_sum) <- c("Broadleaf", "Needleleaf")
  # confusion_mat_spp_sum[,] <- 0
  
  #create empty table for level2
  # confusion_mat_spp_sum <- as.data.frame(matrix(ncol = 6, nrow = 6));
  # colnames(confusion_mat_spp_sum) <- c("BI",  "PI", "SF", "Ta",  "TH", "Wc")
  # rownames(confusion_mat_spp_sum) <- c("BI",  "PI", "SF", "Ta",  "TH", "Wc")
  # confusion_mat_spp_sum[,] <- 0
  
  for (i in 1:100){
    library(sampling);  library(randomForest);     library(snow);          library(doParallel)
    
    # c1 <- makeCluster(detectCores() - 1) # convention to leave 1 core for OS
    # registerDoParallel(c1)
    set.seed(i+100)
    
    as.character(input$SPECIES)
    nbclass.sp <- table(input$SPECIES)
    nmin.sp<- min(nbclass.sp)
    tree.sampsize.sp <- rep(nmin.sp, nrow(nbclass.sp))
    
    #SP.rf_gen_3DINT_N <- randomForest(SP.form, input, ntree = 501, mtry = 3, strata = input$Broad_Needle, sampsize = tree.sampsize.sp, na.action = na.roughfix, proximity = TRUE) 
    #SP.rf_gen_3DINT_N <- randomForest(SP.form, input, ntree = 501, mtry = 3, strata = input$Level2, sampsize = tree.sampsize.sp, na.action = na.roughfix, proximity = TRUE)
    SP.rf_gen_3DINT_N <- randomForest(SP.form, input, ntree = 501, mtry = 3, strata = input$SPECIES, sampsize = tree.sampsize.sp, na.action = na.roughfix, proximity = TRUE)
    
    #include deadtrees, no balance sampling
    #SP.rf_gen_3DINT_N <- randomForest(SP.form, input, ntree = 501, mtry = 3,  na.action = na.roughfix, proximity = TRUE)
    
    
    confusion_mat_spp <- as.data.frame(SP.rf_gen_3DINT_N$confusion)
    confusion_mat_spp_2 <-  confusion_mat_spp[1:(length(confusion_mat_spp)-1)]
    confusion_mat_spp_sum <- confusion_mat_spp_sum + confusion_mat_spp_2
    
    rowsum_species    <- rowSums(as.matrix(confusion_mat_spp[1:(length(confusion_mat_spp)-1)]))
    total_n_err_spp   <- confusion_mat_spp$class.error * rowsum_species
    OOB_error         <- sum(total_n_err_spp) / sum(rowsum_species)
    OOB_error_rf      <- c(OOB_error_rf, OOB_error)
    total_n_corr_spp   <- rowsum_species - total_n_err_spp
    Accuracy_rf_each  <- sum(total_n_corr_spp) / sum(rowsum_species)
    Accuracy_rf      <- c(Accuracy_rf, Accuracy_rf_each)
    #cat(i, "OOB_error_rf", OOB_error, " AND", "Accuracy_rf", Accuracy_rf_each, "\n")
    
  }
  # return(list( Classification_accuracy = data.frame(OOB_error = OOB_error_rf, Accuracy = Accuracy_rf), 
  #              OOB_error_mean = mean(OOB_error_rf), 
  #              Accuracy_mean = mean(Accuracy_rf)))
  return(list( Classification_accuracy = data.frame(OOB_error = OOB_error_rf, Accuracy = Accuracy_rf),
               OOB_error_mean = mean(OOB_error_rf),
               Accuracy_mean = mean(Accuracy_rf),
               confusion_mat_spp_sum = confusion_mat_spp_sum))
  # stopCluster(c1)
  # registerDoSEQ()
  
}